Make no mistake: Enterprise AI is big business, especially for IBM.
IBM Already has $2 billion in generative AI business and is now seeking to speed up that growth. IBM is expanding its enterprise AI business today with the launch of the third generation of Granite Large Language Models (LLMs). A core element of the brand new generation is the continued give attention to true open source enterprise AI. IBM goes one step further and ensures that models might be optimized for enterprise AI with its InstructLab capabilities.
The recent models announced today include general-purpose options with a 2B and an 8B Granite 3.0 model. There are also Mixture of Experts (MoE) models which include Granite 3.0 3B A800M Instruct, Granite 3.0 1B A400M Instruct, Granite 3.0 3B A800M Base and Granite 3.0 1B A400M Base. To round out the update, IBM also has a brand new group of optimized guardrail and safety options, including the Granite Guardian 3.0 8B and Granite Guardian 3.0 2B models. The recent models will probably be available on IBM's watsonX service, in addition to Amazon Bedrock, Amazon Sagemaker and Hugging Face.
“As we mentioned in our last earnings call, the business we’ve built on generative AI is now over $2 billion in technology and consulting,” said Rob Thomas, senior vp and chief business officer at IBM, during a press conference briefing with press and analysts. “When I take into consideration my 25 years at IBM, I’m undecided we’ve ever had an organization that grew so quickly.”
How IBM desires to advance enterprise AI with Granite 3.0
Granite 3.0 introduces a variety of sophisticated AI models tailored for enterprise applications.
IBM expects the brand new models will help support a variety of enterprise use cases, including customer support, IT automation, business process outsourcing (BPO), application development and cybersecurity.
The recent Granite 3.0 models were trained by IBM's centralized Data Model Factory team, which is liable for sourcing and curating the information used for training.
Dario Gil, senior vp and director of IBM Research, explained that the training process involved 12 trillion data tokens, including each voice data across multiple languages ​​and code data. He emphasized that the most important differences from previous generations are the standard of the information and the architectural innovations utilized in the training process.
Thomas added that it is usually essential to acknowledge where the information comes from.
“Part of our advantage in constructing models is the datasets which might be unique to IBM,” Thomas said. “I’d say we’ve a novel position within the industry where we’re the primary customer for all the pieces we construct, which also gives us a bonus when it comes to the way in which we engineer the models.”
IBM reports high performance benchmarks for Granite 3.0
According to Gil, the Granite models have achieved remarkable results on a wide range of tasks, outperforming the most recent versions of models from Google, Anthropic and others.
“What you see listed here are incredibly powerful models, absolutely state-of-the-art, and we’re very happy with that,” said Gil.
But it's not only raw performance that sets Granite apart. IBM has also placed a powerful emphasis on security and trust, developing advanced “Guardian” models to stop its core models from being jailbroken or producing malicious content. The different model size options are also a vital element.
“We care a lot about this, and what we’ve learned from scaling AI, is that inference cost is critical,” Gil noted. “That’s why we’re so focused on the dimensions of the model category, since it has the combo of performance and inference cost that may be very attractive for scaling use cases across the enterprise.”
Why true open source matters for enterprise AI
A key differentiator for Granite 3.0 is IBM's decision to release the models under the Apache 2.0 open source license approved by the Open Source Initiative (OSI).
There are many other open models, resembling: B. Metas Llama, in the marketplace, which are literally not available under an OSI-approved license. This is a distinction that is vital for some firms.
“We decided that we were going to be absolutely clean on this regard and selected an Apache 2 license in order that we could give our enterprise partners maximum flexibility to do what they should do with the technology,” Gil explained .
The permissive Apache 2.0 license allows IBM partners to construct their very own brands and mental property on the Granite models. This helps foster a sturdy ecosystem of solutions and applications built on Granite technology.
“It completely changes the concept of ​​how quickly firms can adopt AI when you’ve a permissive license that allows contribution, enables community and ultimately enables widespread distribution,” Thomas said.
Looking beyond generative AI to generative computing
Looking ahead, IBM is eager about the following big paradigm shift, which Gil called generative computing.
Essentially, generative computing refers to the power to program computers by providing examples or prompts relatively than explicitly writing step-by-step instructions. This is in step with the capabilities of LLMs like Granite, which might generate text, code and other output based on the input received.
“This paradigm, where we don’t write the instructions but program the pc using examples, is key, and we’re just beginning to feel what that seems like through interaction with LLMs,” Gil said. “You will see us investing and moving very aggressively in a direction where with this paradigm of generative computing we are going to give you the option to implement the following generation of models, agent frameworks and way more, it’s a fundamental one Innovation.” Way to program computers as a consequence of the genetic AI revolution.”
